Announcing OCI Compute A4 Acceleron Instances for AI and Cloud-Native Workloads

Oracle Cloud Infrastructure (OCI) continues to expand its portfolio of high-performance, cost-efficient compute offerings with the introduction of A4 Standard Acceleron, the next evolution of A4 Standard—combining the power of AmpereOne M processors with Oracle’s purpose-built Acceleron architecture.

Designed to meet the rapidly evolving demands of modern cloud applications and AI inference workloads, A4 Standard Acceleron advances OCI’s Arm-based compute platform by delivering consistent performance, improved efficiency, and enhanced scalability for enterprise workloads.

Oracle Acceleron SmartNIC for modern workload performance

A4 Standard Acceleron builds on Oracle Acceleron architecture, bringing together innovations across networking, hardware acceleration, and host integration to optimize workload performance. A key component is the Oracle Acceleron SmartNIC, which increases throughput efficiency at both the core and instance level while maintaining strong isolation, built-in encryption, and seamless, no-downtime updates. 

Oracle Acceleron SmartNIC also enables NVMe-based storage acceleration, delivering faster and more efficient data access without the overhead and complexity of traditional dual-NIC designs. The result is a streamlined infrastructure that enhances performance, strengthens security, and simplifies operations.

A4 Standard Acceleron: Built for Scale-Out and Microservices Workloads

A4 Standard Acceleron is available in both bare-metal and virtual-machine configurations. A4 Standard Acceleron Flex VM instances give customers the flexibility to align compute and memory resources more closely with workload needs, from lightweight containerized applications to enterprise workloads that require higher per-core performance and larger memory footprints.

A4 Standard Acceleron Compute offers flexible configurations to meet diverse workload requirements:

• Bare Metal: Direct access to hardware for maximum performance, with 48 Ampere One-M OCPUs (1 OCPU=2 cores), 768 GB memory, and 100 Gbps networking.
• Virtual Machines (Flex): Customizable VM shapes from 1 to 46 OCPUs and up to 700 GB memory, with 1 to 100 Gbps networking for precise rightsizing.
• Competitively priced at $0.0190 per OCPU-hour and $0.0084 GB-hour.

Advancing OCI’s A4 Platform for Modern Workloads

As organizations scale cloud infrastructure to support distributed applications, real-time analytics, and AI-driven services, they need compute platforms that balance performance, predictability, and cost efficiency.

OCI’s A4 instances, powered by Ampere processors, have established a strong foundation for cloud-native workloads. With Oracle Acceleron-enabled A4 Standard Acceleron, Oracle further enhances this platform by improving efficiency and performance at the infrastructure layer.

A4 Standard Acceleron is built for modern distributed applications and AI inference workloads, including generative AI, agentic systems, computer vision, recommender systems, and natural language processing. It is especially well-suited for environments where model density, predictable latency, and cost efficiency are critical. A4 instances support a broad range of inference scenarios—from small- and medium-sized language models (SLMs) to general machine learning workloads—enabling organizations to optimize both performance and cost.

To see A4 in action, explore the demo highlight reel showcasing real-world AI applications—from intelligent agents to multimodal vision and LLM use cases—highlighting the value of the A4 platform in production environments. Learn more about new OCI Compute A4 Acceleron instances through Ampere’s product launch blog.

Get started with OCI Compute A4 Acceleron Instances

A4 Standard Acceleron instances will begin rolling out across select regions today. To learn more, visit the OCI Compute page or contact your Oracle representative.

Learn more about the full portfolio of new Oracle Acceleron Instances: Introducing the Next Generation of OCI Compute Shapes